Yea GPT also suggested the counter idea when I was trying to figure this out. May have to do that to see what’s going on. I have 3 cameras setup with the automation blueprint. One is a doorbell cam that triggers probably 1-5 times a day. The other is a garage cam which triggers probably about the same, and the other is the bedroom cam I mentioned, which I have set to run 8pm-6am. This one maybe triggers 10x a day. They are all set to person detected, not motion. Looking at my LLMV timeline cal, I see about 20-30 events a day. Even if the event description and title generation count as an event, it’s still way under the 200.
Previously I had 1.5 configured and never had an issue. Automation fired whenever a person was detected. When I noticed the automation stopped working, I found out 1.5 had been deprecated, but 2.0 was supported and even had 150 more RPD than 1.5. So at 50 RPD with 1.5, no issue. With 200 RPD in 2.0, literally nothing changed or added, I hit 200. And what’s also confusing is that it hasn’t reset today. So no events yesterday or today, and I’m still showing at 200 max for today.
Has anyone experienced OpenAI providing lots of false responses? I am asking if there is a blue recycling bin or green garbage can (2 separate scripts) on the left side of my driveway and OpenAI says yes most of the time, even though there is nothing there but grass.
That’s rough, and that issue is over a week old. Too worried to get this running for the moment, IDK if google counts repeated API calls that return 429 towards billing or not, but i’d rather not find out.
Been reading through all the Git issues trying to get this figured out with Gemini to no avail. The option to remove title in the blueprint was removed a while back I guess, and I couldn’t get it to work with generate_title: false in the automation. Tried downgrading the blueprint and integration to 1.42, that didn’t work. For whatever reason, ever since I had to upgrade to Gemini 2.0, It’s running like 10+ requests per event, getting to 200 max very quickly. I don’t know how, with 1.5’s 50 RPD, and anywhere from 20-150 event firings in a day, I never went over once. When I went back to LLMV 1.5, I saw there was a fix for the endless retry loop, but it’s not working for me no matter what I do. If anyone has gotten this figured out, would love to hear how!
After the 1.5 update my Gemini/Google provider settings return “invalid API key” despite using the identical setup for quite some time before. The API key in question is used in other tools right now, so I know it’s valid. I am contemplating downgrading, but curious if anyone had a similar experience?
Came back a day later and it is working again. Patience was the key?
I get good responses if I give it a camera entity to pull the image from.
But if I instead have it analyze an image file the response makes no sense.
It says something like “here are the bounding box detections” and that’s it. Nothing intelligent
Has anyone else been having really varied response times from Gemini lately? They were nearly instant a few weeks back now I’m seeing delays, sometimes of minutes 5-15+ until I get a response back. Tried a bunch of models to see if any makes a difference, really not seeing a difference. I assume it’s API load related but odd that it started soon after recent HA and LLMV updates.
I am on free teir, but I’m staying within rate limits. Actaully maxing the rate limits return a 4xx error, this isn’t the case. This is a processing speed issue, unless Google secretly added response delays to free teirs. But other then the rate limit changes I haven’t seen anything about response time delays. At least I’m not the only one seeing this. Been trying to wrap my head around the potential pricing if I moved to OpenAi, at least until I solution a local set-up.
